Skip to main content
Server tools are tools that AI Stats runs for you during a model request. The model decides when to call them, AI Stats executes the tool server-side, and the tool result is fed back into the model loop. Use server tools when the model needs current time, web context, fetched page text, or a second model’s advice without adding your own tool executor.
Server tools are in beta. Request shapes and result payloads may evolve as provider support changes.

​
Available tools

ToolTypeUse it for
Datetimegateway:datetimeCurrent date and time in a chosen timezone
Web Searchai-stats:web_searchModel-directed web searches during a request
Web Fetchai-stats:web_fetchFetching and extracting text from specific URLs
Advisorai-stats:advisorConsulting another model mid-generation
Image Generationai-stats:image_generationCreating images from the model tool loop
Apply Patchai-stats:apply_patchReturning validated patch operations from Responses

​
How server tools work

  1. Add one or more server tools to the tools array.
  2. The model decides whether it needs a tool and emits a tool call.
  3. AI Stats intercepts the server-tool call, executes it, and sends the result back to the model.
  4. The model continues and writes the final response.
Server tools can be used alongside your own function tools. AI Stats executes only the server tools; your application still handles regular function tool calls.

​
Quick start

curl https://api.phaseo.app/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5-nano",
    "messages": [
      { "role": "user", "content": "What are the latest AI policy updates this week?" }
    ],
    "tools": [
      { "type": "ai-stats:web_search" },
      { "type": "gateway:datetime", "parameters": { "timezone": "Europe/London" } }
    ]
  }'

​
Combining with function tools

{
  "model": "openai/gpt-5-nano",
  "messages": [
    { "role": "user", "content": "Find recent funding news and save the best item." }
  ],
  "tools": [
    { "type": "ai-stats:web_search", "parameters": { "max_results": 3 } },
    {
      "type": "function",
      "function": {
        "name": "save_research_item",
        "description": "Save a selected research item",
        "parameters": {
          "type": "object",
          "properties": {
            "title": { "type": "string" },
            "url": { "type": "string" }
          },
          "required": ["title", "url"]
        }
      }
    }
  ]
}
If the model calls ai-stats:web_search, AI Stats executes that call automatically. If it calls save_research_item, your application receives the function call and must return the function result in a follow-up request.

​
Supported endpoints

Server tools are supported on:
  • /v1/chat/completions
  • /v1/responses
  • /v1/messages
Streaming requests are supported. For gateway-managed server tools, AI Stats may materialize the upstream tool-call turn, execute the server tool, continue the loop, and emit a synthetic stream for the final response.

​
Usage tracking

Server-tool usage is reported in usage.server_tool_use: 
{
  "usage": {
    "input_tokens": 105,
    "output_tokens": 250,
    "server_tool_use": {
      "datetime_requests": 1,
      "web_search_requests": 2,
      "web_search_results": 8,
      "web_fetch_requests": 1,
      "advisor_requests": 1,
      "image_generation_requests": 1,
      "apply_patch_requests": 1
    }
  }
}
Pricing can also use server-tool meters such as server_tool_web_search_requests, server_tool_web_search_extra_results, server_tool_web_fetch_requests, server_tool_advisor_requests, server_tool_image_generation_requests, server_tool_apply_patch_requests, native_web_search_requests, and native_web_fetch_requests when configured on model price cards.

​
Tool pages

Datetime

Give the model the current date and time.

Web Search

Let the model search the web when it needs current information.

Web Fetch

Let the model fetch and read specific URLs.

Advisor

Let one model consult another model for guidance mid-generation.

Image Generation

Let the model create images while answering.

Apply Patch

Let the model return patch operations for a client to apply.
Last modified on June 11, 2026